Methods to investigate bacterial stress responses

Given that stress conditions either decelerate bacterial growth or affect the survival of bacterial cells, stress tolerance and resistance are measured by growth ability and survival under stress. Studying the underlying genetic mechanisms of stress responses requires the investigation and specification of a bacterial stress phenotype and linking it to genetic data.

Phenotypic studies on growth and survival

Survival from lethal stressors can be quantified by log10-reductions of bacterial viable counts (Ben Embarek & Huss, 1993; Lundén et al., 2008; Skandamis et al., 2008) and indirectly via minimum inhibitory concentrations (Firsov et al., 1997; Aase et al., 2000; Soumet et al., 2005; Cebrián et al., 2014; Ebner R. et al., 2015). Studies reporting thermal death times also utilize D-values and z-values (Ben Embarek & Huss, 1993; Doyle et al., 2001) that correspond, respectively, to the time in minutes required for a temperature to kill 90%

(log10-cycle) of the population, and the temperature increase in degrees

25

required for a 10-fold (log10) reduction of the D-value. Minimum and maximum growth temperatures also describe tolerance towards temperature stress (Junttila et al., 1988; Hinderink et al., 2009; Markkula et al., 2012a).

Traditionally, the measurement of bacterial growth utilizes viable cell counts from cultures monitored over time, allowing for visualization of growth curves. A downward-sloping logarithmic inactivation curve follows exposure to lethal stress, while susceptibility to mild stress can be estimated by increases in lag phase, decreases in growth rate, and maximum growth level of the sigmoidal growth curve. When mathematically modeling growth under stress, tolerance can be quantified by growth parameters (Jason, 1983; Gibson et al., 1987; Zwietering et al., 1990; Baranyi & Roberts, 1994; Baranyi & Roberts, 1995; Buchanan et al., 1997; Peleg & Corradini, 2011; Huang, 2013; Esser et al., 2015). Common kinetic parameters comprise the following: lag time (lag phase, λ) representing the time period before the beginning of growth;

maximum specific growth rate (growth rate, μ) derived from the slope of the logarithmic growth curve; asymptotic growth level (often abbreviated as “A”) as the maximum level the population reaches by the stationary phase; and area under the curve (AUC) corresponding to the surface area below the growth curve (Korkeala et al., 1992; Firsov et al., 1997; Kahm et al., 2010; Peleg &

Corradini, 2011). As some models assume parameter correlations or do not fit certain types of data, the selection of a suitable parameter estimation approach for each study has been emphasized (Zwietering et al., 1990; López et al., 2004; Peleg & Corradini, 2011; Pla et al., 2015).

The laborious nature of viable cell counts has led to the use of alternative technologies in growth experiments such as the widely utilized turbidity via absorbance (optical density, OD) measurements (Koch, 1970; Korkeala et al., 1992; Sokolovic et al., 1993; Augustin et al., 1999; Faleiro et al., 2003;

Magalhães et al., 2016; Hingston et al., 2017; Keto-Timonen et al., 2018).

Optical density is measured by light transmitted through a sample, where the detection of turbidity depends upon equipment sensitivity and cell densities.

Multiple scattering of light occurs at high cell densities, which increases the probability of light beams detected, and therefore, may underestimate turbidity (Koch, 1970; Stevenson et al., 2016). Additionally, the delayed detection of bacterial growth at low cell densities by OD measuring equipment may lead to inaccurate lag time estimations (Dalgaard & Koutsoumanis, 2001).

It is noteworthy that an apparent increase in cell size without an increase in cell numbers may increase the OD in high stress conditions (Stevenson et al., 2016). L. monocytogenes cells under stress may elongate to form filaments that are divided by septa (Jørgensen et al., 1995; Zaika & Fanelli, 2003;

Hazeleger et al., 2006) and consist of several normal-sized cells on the verge of division (Hazeleger et al., 2006). Consequently, a potential increase in OD caused by these L. monocytogenes filaments corresponds to cell numbers.

Kinetic parameters calculated from OD measurements have been shown to systematically deviate from parameters obtained using viable counts (Dalgaard et al., 1994). Thereby, comparison with traditional viable counts

26

and calibration of OD measurements have been performed (McClure et al., 1993; Dalgaard et al., 1994; Augustin et al., 1999; Dalgaard & Koutsoumanis, 2001; Francois et al., 2005; Pla et al., 2015; Stevenson et al., 2016).

Conversely, if research focuses on the comparison of growth patterns, i.e., growth parameters of several strains relative to one another instead of their absolute values, similar results can be obtained by viable cell counts and OD measurements (Horáková et al., 2004; Pla et al., 2015).

As illustrated in this section, several methodological choices and considerations must be made to perform bacterial growth and stress tolerance experiments. Additionally, strain variability of growth ability should be differentiated from biological variability within individual strains and technical variability within experiments (Aryani et al., 2015). However, execution of the entire data collection and analysis protocol and its repercussions on the comparability of bacterial growth experiments and reliability of stress tolerance studies have not been widely discussed in publications.

Identification of stress-related genetic mechanisms

Common approaches used in the identification of bacterial genetic mechanisms may roughly be categorized as follows: gene expression, genetic modification, and genomic analyses. Depending on the methodology, the level of evidence achieved for the genotype-phenotype interaction is association or causality, the latter of which can be accomplished by fulfilling the molecular Koch’s postulates (Falkow, 2004). In the case of stress responses, these postulates could be modified to read: (i) the investigated stress phenotype should be associated with tolerant/resistant strains, and the genetic trait in question should be present in tolerant/resistant strains but absent in other strains of the bacterial species; (ii) inactivation/removal of the genetic trait should result in a quantifiable loss of stress tolerance/resistence; and (iii) introduction of the genetic trait should result in stress tolerance/resistence.

The expression of several L. monocytogenes genes has been associated with temperature, osmotic, and pH stress by transcriptomic and proteomic analyses (Sokolovic et al., 1993; Bayles et al., 1996; van der Veen et al., 2007;

Chan et al., 2008; Schmid et al., 2009; Mattila et al., 2011; Soni et al., 2011;

Markkula et al., 2012a; Pöntinen et al., 2015). These analyses focus on gene expression by examining RNA transcribed or protein translated upon exposure to an investigated stressor. Conversely, methodologies of genetic modification, such as insertional mutagenesis (Camilli et al., 1990), have associated specific or random inactivated genetic loci with an altered L.

monocytogenes stress tolerance phenotype (Cotter et al., 1999; Kallipolitis &

Ingmer, 2001; Van Der Veen et al., 2009; Hingston et al., 2015). Additionally, deletion mutants combined with the complementation of the identified putatively stress-related genes have confirmed their causality in L.

monocytogenes stress tolerance (Cotter et al., 1999; Dussurget et al., 2002;

27

Kazmierczak et al., 2003; Schmid et al., 2009; Markkula et al., 2012b;

Pöntinen et al., 2017).

The development of DNA sequencing technologies and analysis methods has paved the way to genomic approaches for studying bacterial stress tolerance mechanisms. Genomic comparison of L. monocytogenes strains has yielded associations between particular phenotypes, MLST-types, and the presence or absence of a priori defined stress- and virulence-related genetic loci (Ebner R. et al., 2015; Maury et al., 2016; Hingston et al., 2017).

Conversely, genome-wide association studies (GWAS) provide ways to statistically associate also previously unknown variants derived from a large number of whole-genome sequences with binary or continuous bacterial phenotypes (Farhat et al., 2013; Lees & Bentley, 2016; Lees et al., 2016). In principle, a whole-genome sequence comparison of a closely related stress-resistant and stress-sensitive strain may identify variants present in one but absent in the other, hence indicating an association with the stress response.

However, a large number of strains is required for statistical inference from genome-wide comparisons (Falush & Bowden, 2006; Read & Massey, 2014;

Chen & Shapiro, 2015). Correspondingly, adopting emergent large-scale genome-wide approaches in the search for putative L. monocytogenes stress tolerance loci would first require the generation of reliable high-throughput data on the strain variability of L. monocytogenes stress phenotypes.